Warmup the learning rate

vis_root (string): Root directory of images (e.g. coco/images/) ann_root (string): directory to store the annotation file

vis_root (string): Root directory of images (e.g. coco/images/) ann_root (string): directory to store the annotation file

Download annotation files if necessary. All the vision-language datasets should have annotations of unified format. storage_

Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.

Resume from a checkpoint.

Make causal mask used for bi-directional self-attention.

Load the best checkpoint for evaluation.

Create by split datasets inheriting torch.minigpt_utils.data.Datasets. # build() can be dataset-specific. Overwrite to customize.

Convert applicable model parameters to fp16

Decay the learning rate

Create dataloaders for training and validation.

Given a URL, returns the URL it redirects to or the original URL in case of no indirection